Institute for Telecommunication Sciences / Research / 5G / IORS / 2026.02 IORS

IORS Virtual Meeting, February 25, 2026

AI RAN for 6G

AI is quickly becoming a part of the network. In 5G, AI was largely a feature. With the move to 6G, the fundamental concepts of Open RAN and AI will form the basis of an AI-native network. Testing the operation, resilience, and stability of such AI-driven systems will be critical for future network deployments. This session explored and discussed the past, present, and future approaches to AI testing for Open RAN networks.

Logistics

  • Moderator: Julie Kub (NTIA/ITS)
    • IAN Wong (Viavi Solutions)
    • Charles Turyagyenda (UK SONIC Labs)
    • Eng Wei Koo (Keysight Technologies)
  • Virtual meeting on February 25, 2026 
  • Video recording is available temporarily to registered attendees only

If you have questions, contact us directly at iors@ntia.gov.

Event Summary

AI RAN Testing: Data quality and Model Performance, Viavi Solutions

This presentation introduced VIAVI’s framework for AI‑RAN testing, emphasizing that data quality is essential to ensure reliable AI model performance across RAN applications. Viavi described their Data4AI methodology, which evaluates data integrity, similarity, and diversity to prevent error propagation, ensure dataset task alignment, and reduce deployment risk. Through four real‑world scenarios, Viavi illustrated how the framework validates measurement data, identifies distribution drift, assesses dataset training suitability, and determines when retraining is necessary. 

The second half of the presentation focused on AI model performance testing using a neuromorphic receiver as the device under test. Viavi contrasted white box testing, where model architecture and training information are known, with black box testing, where model details are hidden and test‑case selection must be intelligently optimized. These testing methods provide a scalable, end‑to‑end strategy to ensure trustworthy, generalizable, and energy‑efficient AI models for modern open RAN systems.

AI Native Telecoms: Building the foundations of trust, UK SONIC Labs

UK SONIC Labs (hosted by Digital Catapult) discussed how Artificial Intelligence (AI) is transforming the telecommunications industry, highlighting the importance of trust, verification, and regulatory compliance in deploying AI-driven telecommunication systems. With the shift to AI-native networks, AI will increasingly be embedded in telecommunication systems to manage, optimize, and secure network operations and they mentioned several challenges that must be addressed:

  • Technical: scalability, legacy system integration, cybersecurity, real-time processing
  • Data: quality, bias, privacy, availability of large datasets
  • Ethical and regulatory: transparency, explainability, accountability, trust

The presentation discussed how Standards Development Organizations (SDOs), such as 3GPP, ETSI, O-RAN ALLIANCE, ITU-T, and IEEE, are exploring AI integration into telecommunication systems. However, they need to develop validation and testing frameworks to address:

  • AI for Networks – AI for network planning, optimization, operations, and security
  • Networks for AI – AI infrastructure to run AI systems at scale
  • AI for Customers/Sectors – AI for customer experience and specific applications

The presentation highlighted that a AI lifecycle telecommunication validation framework is critical for: design and use case definition, pre-deployment testing, deployment and operational monitoring, and component retirement or replacement. The framework should be evaluated for key trustworthy AI attributes: accuracy, reliability, robustness, security, fairness, bias mitigation, explainability, transparency, safety, and privacy. Verification methods include black-box testing, which analyses inputs and outputs, and white-box testing, which examines internal model architecture and parameters.

AI RAN Testing with Digital Twin, Keysight Technologies

This presentation emphasized that AI is foundational to 5G-Advanced and 6G, and the industry must prove AI-driven RANs are trustworthy and deployment-ready before going live. AI RAN can be implemented with 5G/6G traditional, virtual, or Open RAN options for terrestrial (TN) and non-terrestrial networks (NTN) using one of three methods:

  • AI-on-RAN: support AI applications in the network
  • AI-for-RAN: use AI applications to improve the network
  • AI-and-RAN: integrate AI into the network fabric 

The presenter noted that there are challenges when moving from rule-based to AI-RAN testing. Data is needed to train AI models. However, sensitive operator data is scarce and often simulated data that replicates the network is required. Testing in realistic emulated environments needs to be performed before deploying solutions to live networks. AI-RAN testing will require unique testing benchmarks and performance gain will need to be weigh against cost and complexity. To successfully deploy AI-RAN, security is a must, including transparency and explainability.

The presentation discussed how moving to an autonomous, AI-native 6G network will require extensive testing to instill confidence in safe, scalable, and trustworthy networks. One way to instill testing confidence is to test using a digital twin that mimics the real network. The digital twin must use trusted, high quality, real world datasets for AI model training and testing. The digital twin must include rigorous lab-based testing and evaluation with precise configuration to model real-world test scenarios.